I would like to see it as extended comments in each git commit. There have been a few examples of some doing so manually but it needs to be supported by the tooling... with all the half-arsed "standards" like MCP etc. I'm surprised there isn't something already.

Internally, we have a standard that any AI written code simply includes a cut-and-paste of the chat prompt (if that were used), and/or the .md files (if those were used).

Good question. The key is sharing the full prompt chain/workflow rather than single prompts - think of it like sharing a Makefile instead of a single shell command. For LLM apps, that means exporting the conversation history, system prompts, and any context files. Tools like Claude Code already support this - you can export full session transcripts. The 'prompt' becomes the reproducible artifact, not just the instruction text.